White Papers

11 Deep Learning Inferencing with Mipsology using Xilinx ALVEO™ on Dell EMC Infrastructure
An ALVEO/Zebra combo installed in a Dell EMC PowerEdge R740 server can be used for
generating high-quality
high-resolution images
from low res images by
mapping a very deep
super-resolution (VDSR)
algorithm onto Zebra.
Super resolution
algorithms are
particularly demanding in
processing power as
they must generate high
resolution images. On a
Dell PowerEdge R740,
Zebra running on ALVEO™
can deliver real time video VDSR movie. The same processing can be applied in video
surveillance to enhance the images that humans have to analyze to identify people.
Applications over internet using image segmentation and classification
Many applications over internet rely on identifying objects, people, text, scenes, activities, body
positioning, or globally the content of images or videos. Relying on all sorts of convolutional
neural networks, their processing requires high bandwidth and flexibility to adapt to the demand.
Dell PowerEdge R740 installed in a data center or a Cloud, equipped with ALVEO running
Zebra can easily process the load. Zebra’s ability to run multiple networks in parallel simplifies
the deployment of applications. The low latency of FPGAs helps reducing the response time of
reactive applications. Transitioning from a farm running inference on CPU/GPU to ALVEO
configured with Zebra is straightforward since Zebra supports the same neural networks without
modifications.
Automation of Quality Control in Manufacturing
An ALVEO/Zebra combo installed in a Dell EMC PowerEdge R740 server can automate quality
control in manufacturing, dramatically increasing the accuracy of the monitoring. By installing
one of them next to the production line, or few of them in a more central location, the quality of
the manufacturing process can increase considerably, and the overall cost noticeably reduced.
Evaluation Methodology
The evaluation was performed via a setup made up by a combination of hardware and software
processing an image-classification application. The hardware accelerator consisted of a Xilinx
ALVEO™ U200 FPGA board hosted by PowerEdge R740/R740xd servers, running Linux.
The software included Mipsology’ s Zebra that computed inference of eight trained deep
learning models or convolutional neural networks running TensorFlow framework.
The image-classification application comprised a complete set of ImageNet images.